Data Augmentation and Correlation Analysis Based on Generative Models

다국어 초록 (Multilingual Abstract)

This study proposes methods for improving the performance of a testbed through data generation and augmentation using generative models, as well as analyzing data correlations and applying them to a recommendation system. The research employs generative models to create and augment datasets to replicate the structure of open datasets. Additionally, by training image-text pairs, the study analyzes relationships between datasets with varying amounts of information and presents an approach for applying this analysis to recommendation systems. The system automatically classifies input images and, based on the classified information, recommends similar images. Thus, this study suggests that the proposed methods for data processing can be applied to multiple domains.

목차 (Table of Contents)

• Ⅰ. 서론 1
• Ⅱ. 배경 2
• 1. Diffusion Model 4
• 2. Vision-based Model 7
• Ⅲ. 본론 12
• 1. Problem Setting 12
• 2. Testbed17
• Ⅳ. 실험 결과 21
• 1. 데이터셋21
• 가. CIFAR데이터셋21
• 나. COCO 데이터셋21
• 다. MUCH 데이터셋 22
• 2. 평가 지표 23
• 3. 실험 결과 25
• 가. Classification25
• 나. Detection26
• 다. Segmentation27
• 라. 정성적 결과 28
• 4. Ablation Study 30
• 가. 실험 환경 30
• 나. 다중 레이블 분류 30
• 다. 실험 결과 32
• Ⅴ. 결론 39
• 참 고 문 헌 40
• ABSTRACT44